Roles
Prompt engineer: designs + evaluates. AI product engineer: integrates + monitors. ML engineer: infra + models. Overlap large.
Advertisement
Review process
Prompt PRs reviewed like code. Include eval results diff. Reject on metric regression. Rubric for prompt clarity + safety.
Advertisement
Eval infrastructure
Shared golden sets. Automated CI eval on PR. Metric dashboards. Alerts on drift.
Documentation
Each prompt: purpose, expected inputs/outputs, model, known limitations, eval results. Living doc in repo.